Ask most people what AI will do to medicine and you'll hear one of two answers: it's going to save millions of lives, or it's going to replace your doctor. Both claims are overstated. The reality developing in hospitals right now is more complicated, and depending on your point of view, more interesting.
The fear is understandable. AI diagnostic tools are genuinely impressive. Systems trained on imaging data now outperform radiologists on specific tasks, particularly spotting certain cancers in mammograms or retinal disease in eye scans. When Google published research showing an AI could read mammograms more accurately than radiologists under certain conditions, the headlines declared radiology a dying profession. Radiology job postings have since increased.
What happened? The tools got deployed, but not as replacements. They became a second set of eyes: flagging cases for human review, prioritizing urgent findings, reducing the chance that something gets missed at 11pm by a tired doctor. Radiologists who use AI tools handle more cases, with better catch rates, than those who don't.
This pattern keeps repeating across specialties. AI systems that analyze pathology slides, flag sepsis risk in ICUs, or suggest drug dosages are now in production at major hospital systems. A 2026 survey by Fierce Healthcare found that three quarters of US health systems are already running AI. In nearly every deployment, the AI assists the clinician rather than replacing them.
The reason is partly technical. Medical AI tools are narrow. A model trained to read chest X-rays does not read ECGs. Neither one takes a patient history, considers drug interactions, factors in the patient's stated preferences, or handles the unexpected complication that doesn't match any training example. Medicine is full of those.
The more immediate change is administrative. Physicians in the US spend, on average, nearly twice as much time on documentation and paperwork as they do on direct patient care. AI tools that draft clinical notes, summarize records, manage prior authorizations, and code diagnoses are already reducing that burden at scale. This is genuinely welcome news for doctors, who consistently report that burnout is driven more by paperwork than by clinical complexity.
The Novo Nordisk-OpenAI partnership announced this week illustrates where the big AI bets in medicine are actually being placed: drug discovery, clinical trial optimization, manufacturing, and supply chain. These are tasks that require pattern-matching across datasets no human could process, and where AI has real advantages. They're also far removed from the patient-facing work that most people picture when they think about a doctor.
Will AI change what doctors do? Yes, substantially. Will it eliminate the profession? That's not where the evidence points. The more likely scenario is that AI takes on more of the routine work, documentation, and pattern recognition, while clinical judgment remains with trained people. That's not a comforting message if you're in a purely administrative healthcare role. For physicians, it's closer to good news.
Commentarii · 0